Improved Geo-Sensing Using Artificial Intelligence Techniques For Tomographic Interpretation

The Bureau of Mines is currently investigating the geophysical application of Tomography to monitor leachate during in-situ mining. Determining the location of the leachate is important for both economic and environmental reasons. By analyzing a sequence of tomographic images, we can determine where highly permeable/fractured regions are located and hence, major flow paths. Since BOMTOM (Bureau Of Mines TOMography - Tomographic Reconstruction Program), and BOMCRATR (Bureau Of Mines Curved RAy Tomography - Tomographic Reconstruction Program) are sensitive to user-selected ray path arrival times, we are also investigating the use of neural networks to automate the process of recognizing this feature. This paper presents a technique for detecting fracture zones and their orientations in a region between two boreholes, by using a motion analysis study of a sequence of tomographic images. It also addresses the need for an automated/unbiased method to pick crosshole seismic arrival times using a trained neural network. In exploration geophysics, the picking of first break arrivals represents a pattern recognition task. The neural net paradigm has proved to be a good method of solving this type of task. To pick the first arrival peak of a trace, the network is trained with a set of known (user selected) first arrival peaks by creating a binary image of a set of traces. Once trained, the neural net correctly picked first arrivals of the remaining traces in the set.